-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathp0lib.f
2174 lines (2174 loc) · 72 KB
/
p0lib.f
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
c-----------------------------------------------------------------------
c basic parallel PIC library for MPI communications
c p0lib.f contains basic communications procedures for 1d partitions:
c NDIAN determines whether number format is big or little endian.
c NDPREC determines whether default reals are double precision.
c IDPREC determines whether default integers are double precision.
c PPINIT initializes parallel processing, returns porcessor ids,
c creates power of 2 sub communicator, and returns its size.
c PPID gets processor ids and number of processors.
c PPINIT0 initializes parallel processing for Fortran77, requests nvp
c processors, and returns processor id
c PPINIT1 initializes parallel processing for Fortran90, returns
c number of processors and processor id.
c PPEXIT terminates parallel processing.
c PPABORT aborts parallel processing.
c HARTBEAT procedure for nodes in main communicator to send a message to
c diagnostic nodes.
c PTIMERA performs synchronized parallel wall clock timing.
c PWTIMERA performs parallel local wall clock timing.
c PSUM performs parallel sum of a real vector, with length a power of 2.
c PISUM performs parallel sum of an integer vector, with length a power
c of 2.
c PCSUM performs parallel sum of a complex vector, with length a power
c of 2.
c PDSUM performs parallel sum of a double precision vector, with length
c a power of 2.
c PMAX performs parallel maximum of a real vector, with length a power
c of 2.
c PIMAX performs parallel maximum of an integer vector, with length a
c power of 2.
c PDMAX performs parallel maximum of a double precision vector, with
c length a power of 2.
c PSCAN performs a parallel prefix reduction of a real vector, with
c length a power of 2.
c PISCAN performs a parallel prefix reduction of an integer vector, with
c length a power of 2.
c PDSCAN performs a parallel prefix reduction of a double precision
c vector, with length a power of 2.
c PBCAST broadcasts real data.
c PBICAST broadcasts integer data.
c PBCCAST broadcasts complex data.
c PBDCAST broadcasts double precision data.
c P0COPY copies real data on node 0 to other nodes.
c PWRITE1 collects distributed real data and writes to a direct access
c binary file.
c PREAD1 reads real data from a direct access binary file and
c distributes it.
c PWRDATA collects distributed real data and writes it to a fortran
c unformatted sequential file.
c PRDDATA reads data from a fortran unformatted sequential file and
c distributes it.
c PWRPART collects distributed particle data and writes it to a fortran
c unformatted sequential file.
c PRDPART reads particle data from a fortran unformatted sequential file
c and distributes it.
c written by viktor k. decyk, ucla
c copyright 1995, regents of the university of california
c update: november 13, 2009
c-----------------------------------------------------------------------
function NDIAN()
c this function determines whether number format is big or little endian
c assumes ieee format for numbers
c ndian = (0,1) = architecture is (little-endian,big-endian)
implicit none
integer NDIAN
integer*4 i
real*8 d
dimension i(2)
equivalence (i,d)
data d /1.0d0/
save
c big endian
if (i(2).eq.0) then
NDIAN = 1
c little endian
else if (i(1).eq.0) then
NDIAN = 0
c error
else
NDIAN = -1
endif
return
end
c-----------------------------------------------------------------------
function NDPREC()
c this subroutine determines whether default reals are double precision
c ndprec = (0,1) = default reals are (normal,double-precision)
implicit none
integer NDPREC
real small, prec, vresult
data small /1.0e-12/
save
prec = 1.0 + small
if (vresult(prec).gt.1.0) then
NDPREC = 1
else
NDPREC = 0
endif
return
end
c-----------------------------------------------------------------------
function IDPREC()
c this subroutine determines whether default integers are double
c precision
c idprec = (0,1) = default integers are (normal,double-precision)
implicit none
integer IDPREC
integer ibig, iprec, iresult
data ibig /2147483647/
save
iprec = ibig + 1
if (iresult(iprec).gt.0.0) then
IDPREC = 1
else
IDPREC = 0
endif
return
end
c-----------------------------------------------------------------------
function vresult(prec)
implicit none
real prec, vresult
vresult = prec
return
end
c-----------------------------------------------------------------------
function iresult(iprec)
implicit none
integer iprec, iresult
iresult = iprec
return
end
c-----------------------------------------------------------------------
subroutine PPINIT(idproc,id0,nvp)
c this subroutine initializes parallel processing
c creates a communicator with number of processors equal to a power of 2
c output: idproc, id0, nvp
c idproc = processor id in lgrp communicator
c id0 = processor id in MPI_COMM_WORLD
c nvp = number of real or virtual processors obtained
implicit none
integer idproc, id0, nvp
c get definition of MPI constants
include 'mpif.h'
c common block for parallel processing
integer nproc, lgrp, lstat, mreal, mint, mcplx, mdouble, lworld
c lstat = length of status array
parameter(lstat=10)
c nproc = number of real or virtual processors obtained
c lgrp = current communicator
c mreal = default datatype for reals
c mint = default datatype for integers
c mcplx = default datatype for complex type
c mdouble = default double precision type
c lworld = MPI_COMM_WORLD communicator
common /PPARMS/ nproc, lgrp, mreal, mint, mcplx, mdouble, lworld
c local data
integer ierror, ndprec, i, k
logical flag
real small, prec, vresult
save /PPARMS/
data small /1.0e-12/
prec = 1.0 + small
c ndprec = (0,1) = (no,yes) use (normal,autodouble) precision
if (vresult(prec).gt.1.0) then
ndprec = 1
else
ndprec = 0
endif
c this segment is used for shared memory computers
c nproc = nvp
c id0 = 0
c idproc = 0
c this segment is used for mpi computers
if (MPI_STATUS_SIZE.gt.lstat) then
write (2,*) ' status size too small, actual/required = ', lstat
1, MPI_STATUS_SIZE
stop
endif
c indicate whether MPI_INIT has been called
call MPI_INITIALIZED(flag,ierror)
if (.not.flag) then
c initialize the MPI execution environment
call MPI_INIT(ierror)
if (ierror.ne.0) stop
c already initialized
else
call PPID(idproc,id0,nvp)
return
endif
lworld = MPI_COMM_WORLD
c determine the rank of the calling process in the communicator
call MPI_COMM_RANK(lworld,id0,ierror)
c determine the size of the group associated with a communicator
call MPI_COMM_SIZE(lworld,nproc,ierror)
c set default datatypes
mint = MPI_INTEGER
mdouble = MPI_DOUBLE_PRECISION
c single precision
if (ndprec.eq.0) then
mreal = MPI_REAL
mcplx = MPI_COMPLEX
c double precision
else
c mint = MPI_INTEGER8
mreal = MPI_DOUBLE_PRECISION
mcplx = MPI_DOUBLE_COMPLEX
endif
c set nvp to largest power of 2 contained in nproc
nvp = 1
idproc = id0
10 i = nvp + nvp
if (i.le.nproc) then
nvp = i
go to 10
endif
c check if nproc is a power of 2
if (nproc.gt.nvp) then
idproc = idproc - 1
endif
c create communicator which contains power of 2 number of processors
c exclude processor 0 in MPI_COMM_WORLD if nproc is not a power of 2
c this processor can be used as a diagnostic node
if ((idproc.ge.0).and.(idproc.lt.nvp)) then
k = 2
else
idproc = MPI_PROC_NULL
k = 1
endif
call MPI_COMM_SPLIT(lworld,k,1,lgrp,ierror)
c requested number of processors not obtained
if (ierror.ne.0) then
write (2,*) ' MPI_COMM_SPLIT error: nvp, nproc=', nvp, nproc
call PPEXIT
stop
endif
c determine the rank of the calling process in the communicator
call MPI_COMM_RANK(lgrp,k,ierror)
if (idproc.ge.0) then
idproc = k
else
idproc = -(k + 1)
endif
c determine the size of the group associated with a communicator
call MPI_COMM_SIZE(lgrp,nproc,ierror)
return
end
c-----------------------------------------------------------------------
subroutine PPID(idproc,id0,nvp)
c this subroutine gets processor ids and number of processors
c output: idproc, id0, nvp
c idproc = processor id in lgrp communicator
c id0 = processor id in MPI_COMM_WORLD
c nvp = number of real or virtual processors obtained
c output: all
implicit none
integer idproc, id0, nvp
c common block for parallel processing
integer nproc, lgrp, mreal, mint, mcplx, mdouble, lworld
c nproc = number of real or virtual processors obtained
c lgrp = current communicator
c lworld = MPI_COMM_WORLD communicator
common /PPARMS/ nproc, lgrp, mreal, mint, mcplx, mdouble, lworld
c local data
integer ierror
c determine the rank of the calling process in the communicator
call MPI_COMM_RANK(lworld,id0,ierror)
c determine the size of the group associated with a communicator
call MPI_COMM_SIZE(lworld,nvp,ierror)
c special case if diagnostic nodes are present
if (nvp.gt.nproc) then
idproc = id0 - 1
if (idproc.ge.nproc) idproc = nproc - idproc - 2
else
idproc = id0
endif
nvp = nproc
return
end
c-----------------------------------------------------------------------
subroutine PPINIT0(idproc,nvp)
c this subroutine initializes parallel processing
c input: nvp, output: idproc
c idproc = processor id in lgrp communicator
c nvp = number of real or virtual processors requested
implicit none
integer idproc, nvp
c get definition of MPI constants
include 'mpif.h'
c common block for parallel processing
integer nproc, lgrp, lstat, mreal, mint, mcplx, mdouble, lworld
c lstat = length of status array
parameter(lstat=10)
c nproc = number of real or virtual processors obtained
c lgrp = current communicator
c mreal = default datatype for reals
c mint = default datatype for integers
c mcplx = default datatype for complex type
c mdouble = default double precision type
c lworld = MPI_COMM_WORLD communicator
common /PPARMS/ nproc, lgrp, mreal, mint, mcplx, mdouble, lworld
c local data
integer ierror, ndprec
logical flag
real small, prec, vresult
save /PPARMS/
data small /1.0e-12/
prec = 1.0 + small
c ndprec = (0,1) = (no,yes) use (normal,autodouble) precision
if (vresult(prec).gt.1.0) then
ndprec = 1
else
ndprec = 0
endif
c this segment is used for shared memory computers
c nproc = nvp
c idproc = 0
c this segment is used for mpi computers
if (MPI_STATUS_SIZE.gt.lstat) then
write (2,*) ' status size too small, actual/required = ', lstat
1, MPI_STATUS_SIZE
stop
endif
c indicate whether MPI_INIT has been called
call MPI_INITIALIZED(flag,ierror)
if (.not.flag) then
c initialize the MPI execution environment
call MPI_INIT(ierror)
if (ierror.ne.0) stop
endif
lworld = MPI_COMM_WORLD
lgrp = lworld
c determine the rank of the calling process in the communicator
call MPI_COMM_RANK(lgrp,idproc,ierror)
c determine the size of the group associated with a communicator
call MPI_COMM_SIZE(lgrp,nproc,ierror)
c set default datatypes
mint = MPI_INTEGER
mdouble = MPI_DOUBLE_PRECISION
c single precision
if (ndprec.eq.0) then
mreal = MPI_REAL
mcplx = MPI_COMPLEX
c double precision
else
c mint = MPI_INTEGER8
mreal = MPI_DOUBLE_PRECISION
mcplx = MPI_DOUBLE_COMPLEX
endif
c requested number of processors not obtained
if (nproc.ne.nvp) then
write (2,*) ' processor number error: nvp, nproc=', nvp, nproc
call ppexit
stop
endif
return
end
c-----------------------------------------------------------------------
subroutine PPINIT1(idproc,id0,nvp)
c this subroutine initializes parallel processing
c output: idproc, id0, nvp
c idproc = processor id in lgrp communicator
c id0 = processor id in MPI_COMM_WORLD
c nvp = number of real or virtual processors obtained
implicit none
integer idproc, id0, nvp
c get definition of MPI constants
include 'mpif.h'
c common block for parallel processing
integer nproc, lgrp, lstat, mreal, mint, mcplx, mdouble, lworld
c lstat = length of status array
parameter(lstat=10)
c nproc = number of real or virtual processors obtained
c lgrp = current communicator
c mreal = default datatype for reals
c mint = default datatype for integers
c mcplx = default datatype for complex type
c mdouble = default double precision type
c lworld = MPI_COMM_WORLD communicator
common /PPARMS/ nproc, lgrp, mreal, mint, mcplx, mdouble, lworld
c local data
integer ierror, ndprec, idprec
integer ibig, iprec, iresult
logical flag
real small, prec, vresult
save /PPARMS/
data small /1.0e-12/
data ibig /2147483647/
prec = 1.0 + small
iprec = ibig + 1
c ndprec = (0,1) = (no,yes) use (normal,autodouble) precision
if (vresult(prec).gt.1.0) then
ndprec = 1
else
ndprec = 0
endif
c idprec = (0,1) = (no,yes) use (normal,autodouble) integer precision
if (iresult(iprec).gt.0) then
idprec = 1
else
idprec = 0
endif
c this segment is used for shared memory computers
c nproc = nvp
c id0 = 0
c idproc = 0
c this segment is used for mpi computers
if (MPI_STATUS_SIZE.gt.lstat) then
write (2,*) ' status size too small, actual/required = ', lstat
1, MPI_STATUS_SIZE
stop
endif
c indicate whether MPI_INIT has been called
call MPI_INITIALIZED(flag,ierror)
if (.not.flag) then
c initialize the MPI execution environment
call MPI_INIT(ierror)
if (ierror.ne.0) stop
endif
lworld = MPI_COMM_WORLD
lgrp = lworld
c determine the rank of the calling process in the communicator
call MPI_COMM_RANK(lgrp,idproc,ierror)
c determine the size of the group associated with a communicator
call MPI_COMM_SIZE(lgrp,nproc,ierror)
c set default datatypes
mint = MPI_INTEGER
mdouble = MPI_DOUBLE_PRECISION
c single precision real
if (ndprec.eq.0) then
mreal = MPI_REAL
mcplx = MPI_COMPLEX
c double precision real
else
mreal = MPI_DOUBLE_PRECISION
mcplx = MPI_DOUBLE_COMPLEX
endif
c single precision integer
c if (idprec.eq.0) then
c mint = MPI_INTEGER
c double precision integer
c else
c mint = MPI_INTEGER8
c endif
nvp = nproc
id0 = idproc
return
end
c-----------------------------------------------------------------------
subroutine PPEXIT
c this subroutine terminates parallel processing
implicit none
c common block for parallel processing
integer nproc, lgrp, mreal, mint, mcplx, mdouble, lworld
c lworld = MPI_COMM_WORLD communicator
common /PPARMS/ nproc, lgrp, mreal, mint, mcplx, mdouble, lworld
integer ierror
logical flag
c indicate whether MPI_INIT has been called
call MPI_INITIALIZED(flag,ierror)
if (flag) then
c synchronize processes
call MPI_BARRIER(lworld,ierror)
c terminate MPI execution environment
call MPI_FINALIZE(ierror)
endif
return
end
c-----------------------------------------------------------------------
subroutine PPABORT
c this subroutine aborts parallel processing
implicit none
c common block for parallel processing
integer nproc, lgrp, mreal, mint, mcplx, mdouble, lworld
c lworld = MPI_COMM_WORLD communicator
common /PPARMS/ nproc, lgrp, mreal, mint, mcplx, mdouble, lworld
integer errorcode, ierror
logical flag
c indicate whether MPI_INIT has been called
call MPI_INITIALIZED(flag,ierror)
if (flag) then
errorcode = 1
c terminate MPI execution environment
call MPI_ABORT(lworld,errorcode,ierror)
endif
return
end
c-----------------------------------------------------------------------
subroutine HARTBEAT(msg,n)
c participating nodes send a message msg to non-participating nodes
c msg = message to send
c n = length of message
c input: msg, n
c output: msg
implicit none
integer n
double precision msg
dimension msg(n)
c common block for parallel processing
integer nproc, lgrp, lstat, mreal, mint, mcplx, mdouble, lworld
c lstat = length of status array
parameter(lstat=10)
c nproc = number of real or virtual processors obtained
c lgrp = current communicator
c mdouble = default double precision type
c lworld = MPI_COMM_WORLD communicator
common /PPARMS/ nproc, lgrp, mreal, mint, mcplx, mdouble, lworld
c local data
integer istatus, nvp, idproc, ierror
dimension istatus(lstat)
c determine the size of the group associated with a communicator
call MPI_COMM_SIZE(lworld,nvp,ierror)
c quit if communicators are the same size
if (nvp.eq.nproc) return
c determine the rank of the calling process in the communicator
call MPI_COMM_RANK(lworld,idproc,ierror)
c send message
if (idproc.eq.0) then
call MPI_RECV(msg,n,mdouble,1,999,lworld,istatus,ierror)
elseif (idproc.eq.1) then
call MPI_SEND(msg,n,mdouble,0,999,lworld,ierror)
endif
c broadcast message to non-participating nodes
if (nvp.gt.(2*nproc)) then
call MPI_BCAST(msg,n,mdouble,0,lgrp,ierror)
endif
return
end
c-----------------------------------------------------------------------
subroutine PTIMERA(icntrl,time,dtime)
c this subroutine performs parallel wall clock timing
c input: icntrl, dtime
c icntrl = (-1,0,1) = (initialize,ignore,read) clock
c clock should be initialized before it is read!
c time = maximum/minimum elapsed time in seconds
c dtime = current time
c written for mpi
implicit none
integer icntrl
real time
double precision dtime
dimension time(2)
c get definition of MPI constants
include 'mpif.h'
c common block for parallel processing
integer nproc, lgrp, mreal, mint, mcplx, mdouble, lworld
c lgrp = current communicator
c mreal = default datatype for reals
common /PPARMS/ nproc, lgrp, mreal, mint, mcplx, mdouble, lworld
c local data
integer ierr
real nclock, ttime
double precision jclock
dimension ttime(2)
c initialize clock
if (icntrl.eq.(-1)) then
call MPI_BARRIER(lgrp,ierr)
dtime = MPI_WTIME()
c read clock and write time difference from last clock initialization
else if (icntrl.eq.1) then
jclock = dtime
dtime = MPI_WTIME()
nclock = real(dtime - jclock)
ttime(1) = nclock
ttime(2) = -nclock
call MPI_ALLREDUCE(ttime,time,2,mreal,MPI_MAX,lgrp,ierr)
time(2) = -time(2)
endif
return
end
c-----------------------------------------------------------------------
subroutine PWTIMERA(icntrl,time,dtime)
c this subroutine performs local wall clock timing
c input: icntrl, dtime
c icntrl = (-1,0,1) = (initialize,ignore,read) clock
c clock should be initialized before it is read!
c time = elapsed time in seconds
c dtime = current time
c written for mpi
implicit none
integer icntrl
real time
double precision dtime
c local data
double precision jclock
double precision MPI_WTIME
external MPI_WTIME
c initialize clock
if (icntrl.eq.(-1)) then
dtime = MPI_WTIME()
c read clock and write time difference from last clock initialization
else if (icntrl.eq.1) then
jclock = dtime
dtime = MPI_WTIME()
time = real(dtime - jclock)
endif
return
end
c-----------------------------------------------------------------------
subroutine PSUM(f,g,nxp,nblok)
c this subroutine performs a parallel sum of a vector, that is:
c f(j,k) = sum over k of f(j,k)
c assumes the number of processors nproc is a power of two.
c the algorithm performs partial sums in binary pairs, as follows:
c first, adjacent processors exchange vectors and sum them. next,
c processors separated by 2 exchange the new vectors and sum them, then
c those separated by 4, up to processors separated by nproc/2. at the
c end, all processors contain the same summation.
c f = input and output data
c g = scratch array
c nxp = number of data values in vector
c nblok = number of data blocks
c written by viktor k. decyk, ucla
implicit none
real f, g
integer nxp, nblok
dimension f(nxp,nblok), g(nxp,nblok)
c common block for parallel processing
integer nproc, lgrp, lstat, mreal, mint, mcplx, mdouble, lworld
c lstat = length of status array
parameter(lstat=10)
c nproc = number of real or virtual processors obtained
c lgrp = current communicator
c mreal = default datatype for reals
common /PPARMS/ nproc, lgrp, mreal, mint, mcplx, mdouble, lworld
c local data
integer istatus, ierr, msid
integer idproc, kstrt, ks, l, kxs, k, kb, lb, j
dimension istatus(lstat)
c find processor id
c this line is used for shared memory computers
c idproc = 0
c this line is used for mpi computers
call MPI_COMM_RANK(lgrp,idproc,ierr)
kstrt = idproc + 1
if (kstrt.gt.nproc) return
ks = kstrt - 2
l = 1
kxs = 1
c main iteration loop
10 if (kxs.ge.nproc) go to 60
c shift data
do 30 k = 1, nblok
kb = k + ks
lb = kb/kxs
kb = kb + 1
lb = lb - 2*(lb/2)
c this loop is used for shared memory computers
c do 20 j = 1, nxp
c if (lb.eq.0) then
c g(j,k) = f(j,kb+kxs)
c else
c g(j,k) = f(j,kb-kxs)
c endif
c 20 continue
c this segment is used for mpi computers
if (lb.eq.0) then
call MPI_IRECV(g,nxp,mreal,kb+kxs-1,l+nxp,lgrp,msid,ierr)
call MPI_SEND(f,nxp,mreal,kb+kxs-1,l+nxp,lgrp,ierr)
else
call MPI_IRECV(g,nxp,mreal,kb-kxs-1,l+nxp,lgrp,msid,ierr)
call MPI_SEND(f,nxp,mreal,kb-kxs-1,l+nxp,lgrp,ierr)
endif
call MPI_WAIT(msid,istatus,ierr)
30 continue
c perform sum
do 50 k = 1, nblok
do 40 j = 1, nxp
f(j,k) = f(j,k) + g(j,k)
40 continue
50 continue
l = l + 1
kxs = kxs + kxs
go to 10
60 return
end
c-----------------------------------------------------------------------
subroutine PISUM(if,ig,nxp,nblok)
c this subroutine performs a parallel sum of a vector, that is:
c if(j,k) = sum over k of if(j,k)
c assumes the number of processors nproc is a power of two.
c the algorithm performs partial sums in binary pairs, as follows:
c first, adjacent processors exchange vectors and sum them. next,
c processors separated by 2 exchange the new vectors and sum them, then
c those separated by 4, up to processors separated by nproc/2. at the
c end, all processors contain the same summation.
c if = input and output integer data
c ig = scratch integer array
c nxp = number of data values in vector
c nblok = number of data blocks
c written by viktor k. decyk, ucla
implicit none
integer if, ig, nxp, nblok
dimension if(nxp,nblok), ig(nxp,nblok)
c common block for parallel processing
integer nproc, lgrp, lstat, mreal, mint, mcplx, mdouble, lworld
c lstat = length of status array
parameter(lstat=10)
c nproc = number of real or virtual processors obtained
c lgrp = current communicator
c mint = default datatype for integers
common /PPARMS/ nproc, lgrp, mreal, mint, mcplx, mdouble, lworld
c local data
integer istatus, ierr, nsid
integer idproc, kstrt, ks, l, kxs, k, kb, lb, j
dimension istatus(lstat)
c find processor id
c this line is used for shared memory computers
c idproc = 0
c this line is used for mpi computers
call MPI_COMM_RANK(lgrp,idproc,ierr)
kstrt = idproc + 1
if (kstrt.gt.nproc) return
ks = kstrt - 2
l = 1
kxs = 1
c main iteration loop
10 if (kxs.ge.nproc) go to 60
c shift data
do 30 k = 1, nblok
kb = k + ks
lb = kb/kxs
kb = kb + 1
lb = lb - 2*(lb/2)
c this loop is used for shared memory computers
c do 20 j = 1, nxp
c if (lb.eq.0) then
c ig(j,k) = if(j,kb+kxs)
c else
c ig(j,k) = if(j,kb-kxs)
c endif
c 20 continue
c this segment is used for mpi computers
if (lb.eq.0) then
call MPI_ISEND(if,nxp,mint,kb+kxs-1,l+nxp,lgrp,nsid,ierr)
call MPI_RECV(ig,nxp,mint,kb+kxs-1,l+nxp,lgrp,istatus,ierr)
else
call MPI_ISEND(if,nxp,mint,kb-kxs-1,l+nxp,lgrp,nsid,ierr)
call MPI_RECV(ig,nxp,mint,kb-kxs-1,l+nxp,lgrp,istatus,ierr)
endif
call MPI_WAIT(nsid,istatus,ierr)
30 continue
c perform sum
do 50 k = 1, nblok
do 40 j = 1, nxp
if(j,k) = if(j,k) + ig(j,k)
40 continue
50 continue
l = l + 1
kxs = kxs + kxs
go to 10
60 return
end
c-----------------------------------------------------------------------
subroutine PCSUM(f,g,nxp,nblok)
c this subroutine performs a parallel sum of a vector, that is:
c f(j,k) = sum over k of f(j,k)
c assumes the number of processors nproc is a power of two.
c the algorithm performs partial sums in binary pairs, as follows:
c first, adjacent processors exchange vectors and sum them. next,
c processors separated by 2 exchange the new vectors and sum them, then
c those separated by 4, up to processors separated by nproc/2. at the
c end, all processors contain the same summation.
c f = input and output complex data
c g = scratch complex array
c nxp = number of data values in vector
c nblok = number of data blocks
c written by viktor k. decyk, ucla
implicit none
complex f, g
integer nxp, nblok
dimension f(nxp,nblok), g(nxp,nblok)
c common block for parallel processing
integer nproc, lgrp, lstat, mreal, mint, mcplx, mdouble, lworld
c lstat = length of status array
parameter(lstat=10)
c nproc = number of real or virtual processors obtained
c lgrp = current communicator
c mreal = default datatype for reals
common /PPARMS/ nproc, lgrp, mreal, mint, mcplx, mdouble, lworld
c local data
integer istatus, ierr, msid
integer idproc, kstrt, ks, l, kxs, k, kb, lb, j
dimension istatus(lstat)
c find processor id
c this line is used for shared memory computers
c idproc = 0
c this line is used for mpi computers
call MPI_COMM_RANK(lgrp,idproc,ierr)
kstrt = idproc + 1
if (kstrt.gt.nproc) return
ks = kstrt - 2
l = 1
kxs = 1
c main iteration loop
10 if (kxs.ge.nproc) go to 60
c shift data
do 30 k = 1, nblok
kb = k + ks
lb = kb/kxs
kb = kb + 1
lb = lb - 2*(lb/2)
c this loop is used for shared memory computers
c do 20 j = 1, nxp
c if (lb.eq.0) then
c g(j,k) = f(j,kb+kxs)
c else
c g(j,k) = f(j,kb-kxs)
c endif
c 20 continue
c this segment is used for mpi computers
if (lb.eq.0) then
call MPI_IRECV(g,nxp,mcplx,kb+kxs-1,l+nxp,lgrp,msid,ierr)
call MPI_SEND(f,nxp,mcplx,kb+kxs-1,l+nxp,lgrp,ierr)
else
call MPI_IRECV(g,nxp,mcplx,kb-kxs-1,l+nxp,lgrp,msid,ierr)
call MPI_SEND(f,nxp,mcplx,kb-kxs-1,l+nxp,lgrp,ierr)
endif
call MPI_WAIT(msid,istatus,ierr)
30 continue
c perform sum
do 50 k = 1, nblok
do 40 j = 1, nxp
f(j,k) = f(j,k) + g(j,k)
40 continue
50 continue
l = l + 1
kxs = kxs + kxs
go to 10
60 return
end
c-----------------------------------------------------------------------
subroutine PDSUM(f,g,nxp,nblok)
c this subroutine performs a parallel sum of a vector, that is:
c f(j,k) = sum over k of f(j,k)
c assumes the number of processors nproc is a power of two.
c the algorithm performs partial sums in binary pairs, as follows:
c first, adjacent processors exchange vectors and sum them. next,
c processors separated by 2 exchange the new vectors and sum them, then
c those separated by 4, up to processors separated by nproc/2. at the
c end, all processors contain the same summation.
c f = input and output double precision data
c g = scratch double precision array
c nxp = number of data values in vector
c nblok = number of data blocks
c written by viktor k. decyk, ucla
implicit none
double precision f, g
integer nxp, nblok
dimension f(nxp,nblok), g(nxp,nblok)
c common block for parallel processing
integer nproc, lgrp, lstat, mreal, mint, mcplx, mdouble, lworld
c lstat = length of status array
parameter(lstat=10)
c nproc = number of real or virtual processors obtained
c lgrp = current communicator
c mdouble = default double precision type
common /PPARMS/ nproc, lgrp, mreal, mint, mcplx, mdouble, lworld
c local data
integer istatus, ierr, msid
integer idproc, kstrt, ks, l, kxs, k, kb, lb, j
dimension istatus(lstat)
c find processor id
c this line is used for shared memory computers
c idproc = 0
c this line is used for mpi computers
call MPI_COMM_RANK(lgrp,idproc,ierr)
kstrt = idproc + 1
if (kstrt.gt.nproc) return
ks = kstrt - 2
l = 1
kxs = 1
c main iteration loop
10 if (kxs.ge.nproc) go to 60
c shift data
do 30 k = 1, nblok
kb = k + ks
lb = kb/kxs
kb = kb + 1
lb = lb - 2*(lb/2)
c this loop is used for shared memory computers
c do 20 j = 1, nxp
c if (lb.eq.0) then
c g(j,k) = f(j,kb+kxs)
c else
c g(j,k) = f(j,kb-kxs)
c endif
c 20 continue
c this segment is used for mpi computers
if (lb.eq.0) then
call MPI_IRECV(g,nxp,mdouble,kb+kxs-1,l+nxp,lgrp,msid,ierr)
call MPI_SEND(f,nxp,mdouble,kb+kxs-1,l+nxp,lgrp,ierr)
else
call MPI_IRECV(g,nxp,mdouble,kb-kxs-1,l+nxp,lgrp,msid,ierr)
call MPI_SEND(f,nxp,mdouble,kb-kxs-1,l+nxp,lgrp,ierr)
endif
call MPI_WAIT(msid,istatus,ierr)
30 continue
c perform sum
do 50 k = 1, nblok
do 40 j = 1, nxp
f(j,k) = f(j,k) + g(j,k)
40 continue
50 continue
l = l + 1
kxs = kxs + kxs
go to 10
60 return
end
c-----------------------------------------------------------------------
subroutine PMAX(f,g,nxp,nblok)
c this subroutine finds parallel maximum for each element of a vector
c that is, f(j,k) = maximum as a function of k of f(j,k)
c assumes the number of processors nproc is a power of two.
c the algorithm performs partial sums in binary pairs, as follows:
c first, adjacent processors exchange vectors and sum them. next,
c processors separated by 2 exchange the new vectors and sum them, then
c those separated by 4, up to processors separated by nproc/2. at the
c end, all processors contain the same summation.
c f = input and output data
c g = scratch array
c nxp = number of data values in vector
c nblok = number of data blocks
c written by viktor k. decyk, ucla
implicit none
real f, g
integer nxp, nblok
dimension f(nxp,nblok), g(nxp,nblok)
c common block for parallel processing
integer nproc, lgrp, lstat, mreal, mint, mcplx, mdouble, lworld
c lstat = length of status array
parameter(lstat=10)
c nproc = number of real or virtual processors obtained
c lgrp = current communicator
c mreal = default datatype for reals
common /PPARMS/ nproc, lgrp, mreal, mint, mcplx, mdouble, lworld
c local data
integer istatus, ierr, msid
integer idproc, kstrt, ks, l, kxs, k, kb, lb, j
dimension istatus(lstat)
c find processor id
c this line is used for shared memory computers
c idproc = 0
c this line is used for mpi computers
call MPI_COMM_RANK(lgrp,idproc,ierr)
kstrt = idproc + 1
if (kstrt.gt.nproc) return
ks = kstrt - 2
l = 1
kxs = 1
c main iteration loop
10 if (kxs.ge.nproc) go to 60
c shift data
do 30 k = 1, nblok
kb = k + ks
lb = kb/kxs
kb = kb + 1
lb = lb - 2*(lb/2)
c this loop is used for shared memory computers
c do 20 j = 1, nxp
c if (lb.eq.0) then
c g(j,k) = f(j,kb+kxs)
c else
c g(j,k) = f(j,kb-kxs)
c endif
c 20 continue
c this segment is used for mpi computers
if (lb.eq.0) then
call MPI_IRECV(g,nxp,mreal,kb+kxs-1,l+nxp,lgrp,msid,ierr)
call MPI_SEND(f,nxp,mreal,kb+kxs-1,l+nxp,lgrp,ierr)
else
call MPI_IRECV(g,nxp,mreal,kb-kxs-1,l+nxp,lgrp,msid,ierr)
call MPI_SEND(f,nxp,mreal,kb-kxs-1,l+nxp,lgrp,ierr)
endif
call MPI_WAIT(msid,istatus,ierr)
30 continue
c find maximum
do 50 k = 1, nblok
do 40 j = 1, nxp
f(j,k) = amax1(f(j,k),g(j,k))
40 continue
50 continue
l = l + 1
kxs = kxs + kxs
go to 10
60 return
end
c-----------------------------------------------------------------------
subroutine PIMAX(if,ig,nxp,nblok)
c this subroutine finds parallel maximum for each element of a vector
c that is, if(j,k) = maximum as a function of k of if(j,k)
c assumes the number of processors nproc is a power of two.
c the algorithm performs partial sums in binary pairs, as follows:
c first, adjacent processors exchange vectors and sum them. next,
c processors separated by 2 exchange the new vectors and sum them, then
c those separated by 4, up to processors separated by nproc/2. at the
c end, all processors contain the same summation.
c if = input and output integer data
c ig = scratch integer array
c nxp = number of data values in vector